Comput. Methods Appl. Mech. Engrg.

195 (2006) 6604–6625

www.elsevier.com/locate/cma

Hot and cold strip rolling processes

Pierre Montmitonnet *

Ecole des Mines de Paris, CEMEF, UMR CNRS 7635, BP 207, 06904 Sophia-Antipolis Cedex, France

Received 3 July 2004; received in revised form 18 October 2005; accepted 20 October 2005

Abstract

Based on the peculiarities of strip rolling processes (long steady-state regime, two-scale character—meter/millimetre—due to local
deformation), and on the practical requirement from industry, the models in use for the mechanical study of cold strip rolling are briefly
reviewed: 1D slab method (SM), 2D/3D upper bound (UB), Finite Difference (FDM) and Finite Element (FEM) methods. Particular
emphasis is put on steady state formulations for lower cost, on roll and stand deformation to study the essential problems of profile
and flatness defects. Major open questions are evoked in conclusion.
 2005 Elsevier B.V. All rights reserved.

Keywords: Strip rolling; FEM; Steady-state

1. Introduction

Being the most widespread metal forming process, strip rolling (pictured schematically in Fig. 1) has received intensive
attention from mechanical engineers since the beginning of the previous century: successful attempts at solving simplified
mechanical equations to disclose the mechanics of cold strip rolling can be traced back to 1925 at least [1]. Also, the early
availability of well-established solutions [2] and of experimental results, the relative simplicity of its geometry, made it an
ideal benchmark for nonlinear computational mechanics, so that virtually any possible method or formulation has been
applied to strip rolling processes.
A fundamental feature that made it possible to fully understand the mechanical bases of the process is its quasi-steady-
state character: except for short and often negligible end effects, the geometry (cross-section), the mechanical and thermal
fields (stress r, velocity v, strain rate tensor e_ , tensorial and scalar strain measures, temperature T) remain stable during the
greatest part of the operation, in particular in finishing hot strip rolling and in cold strip rolling.
On the other hand, factors of complexity include:

• the multistage character of strip rolling on tandem mills where the strips is being bitten in several stands at the same
time, transferring tensions and vibrations from a stand to its neighbours (Section 2.4);
• the local character of deformation: a strip hundreds of meters long is being deformed on a contact length of a few mm to
a few tens of mm at a time; however, a consequence of Saint-Venant’s principle, plastic deformation in the roll bite pro-
motes elastic stress fields extending on both sides on distances comparable with the strip width, i.e. meters (Section 3.3.2,
see Fig. 6).

*
Tel.: +33 (0)4 93 95 74 14; fax: +33 (0)4 92 38 97 52.
E-mail address: pierre.montmitonnet@ensmp.fr

0045-7825/$ - see front matter  2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2005.10.014
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6605

Nomenclature

a thermal diffusivity a = k/qC

Bijkl discretized symmetric gradient operator ð_eij ¼ Bijkk  V kk Þ
C specific heat (per unit mass)
E Young’s modulus
E vector of all global forces on all rolls (roll deformation)
e_ strain-rate deviator e_ ¼ e_  13 Trð_eÞ  I
f k(y) inter-roll force distribution on contact line k per unit length along y (roll deformation)
fjk local inter-roll force, at y = yj on contact line k (roll deformation)
F vector of all discretized force distributions on all roll contact lines (roll deformation)
F1, F2 entry, exit tension forces
Ftot total screwdown force (roll deformation)
F krb roll bending force on roll k (roll deformation)
h(x, y) strip thickness; h1, h2: entry, exit thickness
k thermal conductibility
lkc ðyÞ contact arc length on contact line k (roll deformation)
L strip/work roll bite length
m strain rate sensitivity of the yield stress
n vector normal to a surface
Nk FE shape function associated to node k
p hydrostatic pressure p ¼  13 TrðrÞ
pk(x, y) contact pressure distribution on contact line k (roll deformation)
qk(y) net force distribution on roll k, sum of all f ’s on all its contact lines (roll deformation)
qkj local net inter-roll force, at y = yj on roll k (roll deformation)
Q vector of net force distribution on all rolls (roll deformation)
r radial coordinate (thermal, roll)
R(Ra) roll radius (radius of the roll axis)
R residual vector (components Rkk, k for nodes, k for coordinates)
s stress deviator s ¼ r  13 TrðrÞ  I
S surface area (dS surface integration element; S1, S2: entry, exit cross-sectional areas)
t time
tb strip tensions (b = 1 for entry, b = 2 for exit)
T(x, y, z, t) temperature field
u(x, y, z, t) displacement vector (components ui or ux, uy, uz)
v(x, y, z, t) velocity vector (components vi or vx, vy, vz)
V nodal velocity vector (components Vkk, k for nodes, k for coordinates)
w strip width
W roll barrel length along y (roll deformation)
Wa distance between roll barrel end and force application point (roll deformation)
x coordinate vector (components xi or x, y, z: coordinates in the rolling direction (RD), the transverse direction
(TD), the vertical or normal direction (ND))

Greeks   rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 h1  h2 h1  h2 L
a bite angle, a ¼ cos 1  
2R R R
aT volumic dilatation coefficient
vel elastic bulk modulus
d contact penetration distance (d0: penetration tolerance)
dT depth of the roll skin model
D operates an increment
e_ (x, y, z, t) strain rate tensor e_ ¼ 12  ðrvþt rvÞ
 1=2
e_ equivalent strain rate e_ ¼ 23 Re_ pl : e_ pl
t
e cumulated plastic strain e ¼ 0 e_  dt
/(x), u(y) shape functions in the influence function method (roll deformation)
6606 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

U heat flux
kpl plastic multiplier
kel, lel Lamé’s elastic coefficients
l Coulomb’s friction coefficient
m Poisson’s ratio
h angular variable
q volumic mass
r(x, y, z, t) stress tensor
r0 ðe; e_ ; T Þ yield stress
rn normal stress rn = n Æ r Æ n
s friction stress vector s = r Æ n  (n Æ r Æ n)n
x angular velocity
X, oX, dX computational domain, its boundary, volume integration element

Superscripts
core in the roll central area (thermal, roll)
J Zaremba-Jaumann derivative
pl, el plastic (resp. elastic)
int interface (between skin and core, thermal, roll)
skin near surface area (thermal, roll)
* virtual field
a_ material time-derivative of variable a

Subscripts
cond conductive (thermal)
conv convective (thermal)
b bending (roll deformation)
d downstream (the neutral point)
f free surface
fm, fs flattening (roll deformation)
p Poisson term (roll deformation)
r roll
RBM Rigid Body Motion (roll deformation)
T thermal
u upstream (the neutral point)

Both features make 3D rolling modelling a two-scale process where the description must be made at a sub-millimetric
scale in the roll bite (even micrometric for thin foil), whereas the system to consider is several meters long and wide.

Ftot/2 Back-up roll (BUR) Ftot/2

z (ND)

Work roll (WR)

y (TD) Fwrb/2 Fburb/2

x (RD)
w

(Ftot+ Mg)/2 (Ftot+ Mg)/2

Fig. 1. Schematic configuration of strip rolling—M is the mass of the roll stack (symmetry with respect to planes Oxz and Oxy is assumed here).
Fwrb, Fburb: work-roll (back-up roll) bending forces. Ftot: screwdown force.
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6607

If so much has been done for such a long time, what is at stake today, what is the driving force for more research input
into this process development? Of course the demand for ever better mechanical properties, hence microstructures (for
automobile weight gains e.g.), geometric precision, and surface quality, together with an ever increasing productivity,
i.e. rolling speed. These in turn require:

• A precise description of the elastic-(visco)plastic (EVP) deformation of the strip (Section 3), in terms of 1D, 2D or 3D
fields depending on the problem under investigation.
• Accounting for the physically very strong coupling of strip plastic and roll thermo-elastic deformation (Section 4) and its
consequences: profile and flatness defects; as ever thinner and harder strips are being rolled, the coupling is stronger and
stronger, and the numerical methods to catch it must become more robust; moreover, to feed regulation loops based on
continuous measurements, it is necessary to model the effects of a number of actuators installed on the mills to coun-
teract roll deformation.
• A reliable and accurate physical input, in terms of volume properties (strain-hardening function, strain-rate and temper-
ature effects—even for cold rolling) and surface properties (friction and thermal transfer, connected with lubricant films
and superficial transformed layers); these are now tackled with micro–macro modelling, which introduces a third, sub-
micrometric scale.
• Chattering (vibration of the mill e.g. in the 100–300 Hz range) becomes a major trouble as rolling speed increases; its
basic physical understanding and its connection with friction in particular are being addressed by specific models.

2. Literature review: quasi-static modelling of strip deformation

2.1. Viscoplastic (VP), Elastic-Viscoplastic (EVP) or Thermo-EVP?

Models are often built with VP or rigid plastic (RP) behaviour (slightly-compressible rigid-plastic in [3] to avoid diffi-
culties connected with the enforcement of incompressibility in flow formulations). Alternatively, elastic-(visco)plastic con-
stitutive equations are assumed. The latter are much more complex (attention must be paid to stress rate objectivity;
constitutive equations are ODE’s which must be time-integrated for each material point at each stage); the counterpart
is their ability to predict residual stresses, a very important item in cold strip rolling: see Fig. 6 which shows quite different
stress patterns predicted by VP or EP models, with major consequences on post-bite flatness analysis.
Thermal effects are of course most important in hot strip rolling, through thermo-mechanical coupling (influence of tem-
perature T on yield stress), but also through its influence on microstructure. To a certain extent, this is sometimes also true
in cold rolling, where e.g. excessive heat of deformation and friction may bring aluminium strip in its thermally activated
regime where recrystallization may take place, with dramatic consequences on strip mechanical properties.

2.2. Steady-state or time-dependent models?

Although dominated by steady state, rolling also exhibits important transient effects, such as end shapes, only important
in hot slab rolling, or flatness defects—i.e. periodic buckled waves travelling with the product. Vibrations (chattering) are
another departure from steady state. Time-dependent, incremental models are fully justified when the above-mentioned
phenomena are investigated.
In all other cases, steady state formulations are preferable owing to much shorter computing time. The main difficulty is
that the shape of the domain X (and its boundary oX) is not known because of free surfaces (lateral surfaces where spread
occurs, top and bottom surfaces where pre- and post-bite deformation phenomena take place). The determination of these
free surfaces is based on the equation of stationarity of the free surface oXf:
For any point xf of the free surface,
vðxf Þ  nf ¼ 0; ð1Þ
where v is the velocity field and nf the normal to the (free) surface at xf. This equation must be solved under the constraint
that v is the solution of the thermo-mechanical flow problem on domain X. Different iterative techniques have been pro-
posed to solve it since the early 1980s: the streamline method has been used most often, e.g. in [4]; it consists, once a velocity
field v (vx, vy, vz) has been determined on an estimate of X, in displacing the nodes according to
dy dz dx
¼ ¼ . ð2Þ
vy vz vx
(x, y, z) are the coordinates in the Rolling Direction (RD), Transverse Direction (TD) and Normal Direction (ND) respec-
tively. Alternative techniques consist in minimizing a functional over the free surface [5]:
6608 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625
Z
ðv  nÞ2 dS ð3Þ
oXf

or solving a weak form of Eq. (2) using Galerkin or Petrov–Galerkin techniques [6,7]:
Z
ðv  nÞ  N  dS. ð4Þ
oXf

Such steady state formulations can be termed quasi-Eulerian (the mesh is fixed in RD, and nodes move along TD and ND
just to cope with Eq. (2)). They are almost systematically associated with VP or RP behaviours, due to the difficulty of
constructing a steady state model for EVP.
In the case of incremental computations, all three types of formulations have been proposed:

• Eulerian, with a fixed mesh in which matter moves [8];

• Standard Updated Lagrangian, by far the most popular formulation;
• Arbitrary Lagrangian–Eulerian (ALE) [9], generally with operator splitting (first a Lagrangian step, then a projection of
fields on the previous mesh with contact adaptation).

2.3. 1D, 2D or 3D models

When rolling wide, thin material (i.e. if the contact width to bite length ratio w/L is large enough), the problem becomes
mostly 2D, except at edges and for temperature, post-bite stresses, and provided roll thermo-elastic deformation is
neglected. Moreover, if the bite length to thickness ratio L/h is large, the gradients along ND are very small, so that most
fields become 1D to a reasonable approximation (except temperature or residual stress again). This leaves space for models
of widely variable sophistication.

2.3.1. 1D models
1D models where velocity, strain and stress fields depend on x only are based on the slab method [1,10–12], which
neglects shearing, non-diagonal stresses so that everywhere, the principal axes are (RD, TD, ND). In the EP or EVP case,
the problem is reduced to a set of 3 ODE’s in the 3 unknown, principal components of stress tensor r. Further reduction is
possible for RP behaviour, since incompressibility brings about two relations between the 3 components:
ryy ¼ ðrxx þ rzz Þ=2 ðflow ruleÞ ð5aÞ
pffiffiffi
jrxx  rzz j ¼ 2r0 = 3 ðyield criterionÞ ð5bÞ
which allows the mechanical equilibrium along RD to be written in terms of rxx only (see Fig. 2):
 
dðhrxx Þ 2r0 tgh  l
¼ 2 rxx  p ffiffi
ffi . ð6Þ
dx 3 1  l  tgh

dx

σn
τ

h σzz h+dh
R
σ xx

τ
σn θ
α

h1/2
F1 h2/2 F2

x
L

Fig. 2. plane strain Slab Method principle.

 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6609

h(x) is the local strip thickness, r0 is the yield stress, l is Coulomb’s friction coefficient; h is the local angle between the
tangent to the roll and horizontal (h 6 a, the bite angle); in the last term, the top signs are valid upstream the neutral point,
the bottom signs downstream. This equation is integrated generally by a 4th order Runge–Kutta method, starting from
boundary conditions:

F1 F2
rxx ¼ ¼ t1 at inlet, rxx ¼ ¼ t2 at outlet, ð7a; bÞ
S1 S2

where F1 and F2, S1 and S2 are respectively the tension forces and cross-sectional areas at entry (subscript 1) and exit (sub-
script 2). The solution of Eqs. (5)–(7), or of their EVP counterpart, has been the foundation of the understanding of the
mechanics of rolling processes, with the notions of neutral point (where the sliding direction, hence the friction stress vec-
tor, reverse), the ‘‘friction hill’’ with its apex at the neutral point, the estimation of the force and torque by integration
along x. These 1D approaches disclose the effects of most process parameters (roll radius R, tension forces F1 and F2, fric-
tion, yield stress and strain-hardening. . .). Most cold rolling models which run in the R&D centres of metallurgical com-
panies are still of this kind, providing excellent quality/cost ratio. Owing to their simplicity, they are also ideal to initiate
studies on complex couplings (roll deformation, friction models. . .).  h2
 qhffiffiffiffiffiffiffiffi
ffi
1 h2
Also note that, considering the small values of the bite angle a ¼ cos1 1  h12R  R
 RL, an analytical formula
for the normal stress is obtained if strain hardening is neglected:

hðhÞ h pffiffiffiffiffiffiffiffiffiffin  pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffioi

ruzz ¼ ðr0  t1 Þ  exp 2l R=h2  tan1 h R=h2 þ tan1 a  R=h2 ; ð8aÞ
h1
hðhÞ n pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffio
rdzz ¼ ðr0  t2 Þ  exp 2l R=h2  tan1 h  R=h2 ; ð8bÞ
h2
hðhÞ ¼ h2 þ 2Rð1  cos hÞ  h2 þ Rh2 ; x ¼ R  sin h  Rh. ð8cÞ

Superscript ‘u’ stands for upstream, ‘d’ for downstream the neutral point. Although hardly usable for industrial purposes
(absence of workhardening), these standard formulae are a very good didactic tool, explaining most of the major features
of the roll bite mechanics.

2.3.2. 2D models
In cold temper rolling processes (very small thickness reduction), deformation is strongly z-dependent (with a concen-
tration near the strip/roll contact). The same is true in hot strip rolling, in particular in the roughing mill section. In the
finishing section, the strains and stresses are more or less z-independent in the last stands, but the strong temperature cou-
pling brings the z-coordinate back in. In such cases, 1D approaches are clearly inadequate and must be replaced by plane
strain 2D models, whereby fields depend on 2 space variables, namely x and z. Four types of methods give excellent pre-
dictions if properly used:

• upper-bound-like methods, with velocity fields based on streamlines generally, using either VP or RP (slightly compress-
ible) [13];
• Finite Difference Methods (FDM) [14,15];
• of course FEM ([16,17] to quote only pioneering work), generally quasi-static implicit [18], sometimes dynamic explicit
[19];
• and recently, the Element-Free Galerkin method [20].

The last three may use any kind of constitutive equations. Compared with 1D models, 2D ones account for shear strains
and stresses, for z-dependence of velocity, strain, stress and temperature, and therefore are able to predict profiles of resid-
ual stress rxx(z), of temperature and microstructure, or to examine microscopic, superficial phenomena such as iron fines
formation [21] or oxide scale breaking and embedding [22,23]. Moreover, they are able to deal with end effects as well as
steady state, whereas 1D models are confined to the latter.

2.3.3. 3D models
The necessity of 3D models arises when dealing with thick and relatively narrow products (roughing mill), or when con-
sidering the roll deformation and its very important consequences in terms of profile and flatness defects, or residual stres-
ses. The methods include generalized Upper Bound Methods (Finite Strip Methods, with rigid plastic behaviour [24]),
FDM [14,15], and, widely dominant, FEM, again implicit (see [25] among many others) or sometimes dynamic explicit
formulations.
6610 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

v=? v=?

t =? t =?

v12 = v12 v 22 = v13

t 21 = t12 t 22 = t13

Fig. 3. Illustration of the multi-stand coupling algorithm in [26]. Roll velocities are imposed. The unknown inter-stand velocities and tensions are
determined by decoupling the stands and their inlet and outlet velocities and tensions, and writing relevant equalities between them, approached
iteratively.

2.4. Tandem mill modelling

When dealing with tandem mills, interstand tensions must be accounted for due to their significant influence on spread
(mainly for narrow products) or on stress profiles along TD. Early models used to compute one stand at a time, assuming
some value of tension forces F1 and F2, or using measured ones. It was however desirable to predict these forces based on
the setting of the actuators, the roll speeds of the different stands. With 2D FEM models, it is now possible to include all
stands in one and the same simulation, whereby interstand tension fields and their effects come out as any other field. This
remains too heavy in 3D, where [26] proposes a different, iterative strategy including series of single stand simulations
(Fig. 3). It assumes the absence of interstand deformation, which implies that the velocity field V ki;2 in the exit cross-section
of stand k equals the velocity field V kþ1
i;1 in the entry cross-section of stand k + 1 (here, subscript i stands for any node in a
cross-section). Starting from tension estimates tkj;b (with tkj;2 ¼ tkþ1
j;1 ), each stand is modelled separately, giving among other
oV ki;a
results unbalanced V kþ1 k
i;1  V i;2 . Then, the influence coefficient matrix otkj;b
ða; b ¼ 1; 2Þ is inverted to compute new tkj;b ’s
which will eliminate the unbalanced velocities, and a new series of single stand computations is made. This matrix can
be calculated straightforwardly using the current residual vector R and stiffness matrix oR/oV:
 1
dR oR oR oV oV oR oR
8tb ; RðV; tb Þ ¼ 0 ) ¼ þ  ¼0) ¼  ;
dtb otb oV otb otb oV otb
Z Z Z ð9Þ
Rkk ¼ rij  Bijkk  dX  sk  N k  dS  tk  N k  dS.
X oXc oXe

Boundary oXc is the contact surface, oXt is the cross-sections on which tensions are applied. Nk’s are the shape functions,
Bijkk is the discrete symmetrized gradient operator.
Application to tube rolling on Mandrel Pipe Mill is demonstrated in [26].

3. A 3D, EVP, quasi-Eulerian steady-state model

In [27,28], a 3D FEM of hot and cold strip rolling has been described. The behaviour can be either EVP or VP. Two
slightly different steady state formulations have been implemented depending on the behaviour; both are based on the
streamline method. Incremental, non-steady state simulations are also possible; the formulation, being a trivial variant
of the model, will not be described here. In this part, the tools (rolls) are assumed rigid.

3.1. Basic equations and spatial discretization

3.1.1. Mechanical equilibrium equation

Mechanical equilibrium (conservation of momentum) is written in terms of the Principle of Virtual Power (PVP):
Z Z
 
For any kinematically admissible v ; r : e_  dX ¼ ðr  nÞ  v dS. ð10Þ
X oX

e_ is the strain rate tensor (with deviator e_ ). The mass conservation condition must be added (q is the volumic mass):
q_
DivðvÞ ¼  . ð11Þ
q
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6611

As isochoric plasticity is used, the only volume variation is elastic, so that, introducing the elastic compressibility coefficient
el
vel ¼  q1  dq
dp
¼ kel þ 2l3 :
p_ ¼ vel  Trð_eÞ. ð12Þ
el el
k and l are the elastic Lamé coefficients, Tr(Æ) means the Trace operator, a superimposed dot means the material time
derivative.
A flow formulation (with velocity and temperature as the sole nodal unknowns) is chosen. The space discretization of
the PVP by the FEM thus writes, using shape function Nk:
Z Z
For any dofðk; kÞ; Rkk ¼ rij : Bijkk  dX  ðr  nÞk  N k  dS ¼ 0. ð13aÞ
XðtÞ oXðtÞ

Or, introducing the pressure p/deviator s split and specifying the surface terms:
Z Z Z Z
Rkk ¼ sij : B0ijkk  dX  p  Biikk  dX  sk  N k  dS þ tbk  N k  dS ¼ 0; ð13bÞ
XðtÞ XðtÞ oXcðtÞ oXtbðtÞ
 
1 oN k oN k oN k 1
Bijkk ¼   djk þ  dik ; Biikk ¼ ; B0ijkk ¼ Bijkk  Biikk  dij ð13cÞ
2 oxi oxj oxk 3

(k = 1, . . . , number of nodes stands for a node, k = 1, 2, 3 for a space coordinate). s is the friction stress, tb is the interstand
tension stress.
Tri-linear, 8-node hexahedra are used. Integration is performed using 8 Gauss points for the volume and 4 Gauss points
for the surface elements; reduced integration (1 central Gauss point) is necessary for the pressure term to avoid locking and
oscillations.
The nonlinear system of Eq. (13b) in V is solved by a standard Newton–Raphson method with line search.

3.1.2. Constitutive equations

A simulation may involve several materials (sandwich rolling, . . .). Although several constitutive models are available in
the software (including pure VP Norton–Hoff behaviour, and anisotropic VP equations [29], polycrystalline coupling [30]
and other microstructure-coupled models), only the hypo-elastic, Prantdl–Reuss EVP constitutive equations are described
here. They are best written after decomposition of r into the hydrostatic pressure p and stress deviator s:

 
if s : s < 23  r20 ðe; e_ ; T Þ or s : s ¼ 23  r20 ðe; e_ ; T Þ and s : s_ < 0 (elastic state):

p_ ¼ vel  Trð_eÞ; ð12Þ

J el
e_ ¼ s_ =2l . ð14aÞ
 
Else s : s ¼ 23  r20 ðe; e_ ; T Þ and s : s_ > 0 (plastic state):
p_ ¼ vel  Trð_eÞ; ð12Þ
J
s_ 3 e_
e_ ¼ þ s ; ð14bÞ
2lel 2 r0 ðe; e_ ; T Þ
2
s : s ¼  r20 ðe; e_ ; T Þ ð15Þ
3  
pffiffiffi b  npffiffiffi om 
r0 ðe; e_ ; T Þ ¼ 3  exp  ðe þ e0 Þn  K 0 þ K 1  3  e_ . ð16Þ
T
1=2 R
s_ J is the Jaumann rate of s, e_ ¼ ð2=3  e_ pl : e_ pl Þ the equivalent strain rate and e ¼ trajectory
e_  dt the equivalent strain.

3.1.3. Contact and friction

Unilateral contact is assumed (although bilateral is also available):
ðv1  v2 Þ  n 6 0; rn ¼ ðr  nÞ  n 6 0; ½ðv1  v2 Þ  n  rn ¼ 0. ð17a; b; cÞ
v1 is the velocity vector of body 1, v2 of body 2, and normal vector n points from 1 to 2. In fact, the Signorini formulation
above is relaxed by a penalty technique, which applies a normal repulsive force to any node of body 1 which penetrates into
body 2:
þ
F pen ¼ K pen  hdn  d0 i . ð18Þ
6612 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

dn is the normal distance to the surface of body 2, d0 is a penetration tolerance. Kpen is the penalty coefficient, to be chosen
carefully to find a compromise between the precision of non-penetrability enforcement and the conditioning number of the
stiffness matrix.
Friction is modelled by Coulomb’s coefficient, with a limitation by a friction factor (an anisotropic version is also avail-
able [31]):
 
r0 v2  v1
  pffiffiffi  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi .
s ¼ Min l  rn ; m ð19Þ
3 ðK reg  V c Þ2 þ jv2  v1 j2

Note the regularization by a velocity-dependent term which avoids the tedious treatment of the stick-slip threshold con-
dition (tangential analogue of Eq. (18)). Vc is a characteristic sliding velocity which makes the regularization parameter
Kreg non-dimensional; if Kreg is small enough, the regularization does not change significantly the physics of the contact.

3.1.4. Heat transfer equation

Finally, the heat transfer equation is written as usual in weak integral form:
Z Z Z
dT oN k
qC  N k dX  kðrT Þi  dX þ ðUcond þ Uf ÞN k dS ¼ 0. ð20Þ
XðtÞ dt XðtÞ oxi oXðtÞ

The same element (geometry, shape functions, Gauss points) is used as for velocity. k, q, C are respectively the conduct-
ibility, the volumic mass, the specific heat (all three may depend on temperature). Flux boundary conditions involve a lin-
earized conduction through contact surfaces with rolls, and convection + radiation on free surfaces (environment
temperature T1):
Ucond ¼ hcond  ðT  T roll Þ; ð21aÞ
4
Uf ¼ hconv  ðT  T 1 Þ þ rst  er  ðT  T 1 Þ . ð21bÞ
rst is Stephan’s constant, and er is the total emissivity.
As always, choosing the Heat Transfer Coefficient (HTC) hcond is a major difficulty, quite influent on thermal fields in
the strip and rolls [32]. Measurements display widely varying values, between 300 and 180 000 W m2 K1 [33,34]! From a
physical point of view, the HTC results from the thermal flux line constriction due to partial asperity contact, and to heat
transfer resistant films, both oxides and lubricant if any. It therefore depends primarily on pressure which increases the real
contact area, and interface deformation which breaks oxide and let more conductive metal through; models including some
of these internal variables have been proposed [22,26].
Note that severe difficulties are also encountered in the choice of a HTC hconv for coolant/roll contact under the cooling
sprays [35].

3.2. Steady state formulation

3.2.1. Mechanics
The streamline method (Eq. (2)) is used. Joined to the rather low distortions generally encountered in rolling, this trig-
gered the choice of structured mesh, without remeshing. Two difficulties were met however:

• The integration of streamlines (with the derivatives dy/dx and dz/dx taken at the upstream or downstream node, or
averaged) does not respect contact conditions: superficial streamlines may move away from the roll, or cut it [4]. A sig-
nificant amount of work had to be spent on projection algorithms robust enough for simulations of both long and flat
products [27]. Also, a regularization of the internal as well of external nodal positions is performed, so that nodes (and
Gauss points) are aligned along streamlines.
• For VP behaviour, only the integration of e_ into e has to be performed to update the state variables, stresses are post-
processed from the velocity field. However in EVP, stresses must be explicitly determined and included in Eq. (13b)
before resolution. The constitutive equation is a set of Ordinary Differential Equations (ODEs) in time, which has to
be integrated; as time makes no sense in a steady state model, it has to be replaced by a space integration (using cur-
vilinear coordinates along streamlines).

The algorithm (called ELDTH: Eulerian–Lagrangian with Heterogeneous Dt) of a single ‘‘free surface iteration’’ is pre-
sented as part of the general flow-chart of the model in Fig. 4. A ‘‘pseudo-time step’’ is defined for each integration point
(hence ‘‘Heterogeneous Dt’’): it is the time necessary for it to move to the equivalent Gauss point in the next element down-
stream. Once Dtlocal has been determined as a function of v, De ¼ e_  Dtlocal is used to compute the stress increment Dr by
solving the incremental form of Eqs. (12) and (14)–(16) by the radial return technique:
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6613

Fig. 4. Algorithm of the complete model (including the roll thermomechanical model and coupling), with the Heterogeneous Time-Step Eulerian–
Lagrangian strategy (ELDTH) (dotted frame).

Dp ¼ vel  Trð_eÞ  Dt; ð22aÞ

Ds 3 s þ Ds
e_  Dt ¼ el þ  De  ; ð22bÞ
2l 2 De
r0 ; e þ De; T
Dt
 
2 De
ðs þ DsÞ : ðs þ DsÞ ¼  r20 ; e þ De; T . ð22cÞ
3 Dt
Eqs. (22b) and (22c) can be reduced to a single scalar equation in De:
  rffiffiffiffiffiffiffiffiffiffiffi
De el 3
r0 ; e þ De; T þ 3l De ¼ ~s : ~s; ~s ¼ s þ 2lel e_  Dt. ð23Þ
Dt 2
~s is the elastic predictor. Once De has been determined from Eq. (23), s + Ds is deduced by Eq. (22b). Note that Eq. (22b)
involves only the material derivative of the stress; the rotation term of the Jaumann rate is applied to r + Dr using Xas, the
6614 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

skew-symmetric part of the velocity gradient, assuming that it is sufficient to apply the whole incremental rotation instan-
taneously at the end of the increment.
Finally, stress is integrated along the streamline by

rðxnþ1 Þ ¼ rðxn Þ þ Drðrðxn Þ; v; Dtlocal Þ. ð24Þ

Subscript n refers to the nth integration point along a given streamline. The sub-stepping procedure described in Fig. 4 is
continued until the stress field stabilizes. Note that at each ‘‘sub-step’’ (comprising a velocity computation by nonlinear
solution of Eq. (13b)), stress is moved downstream by one element; the number of sub-steps for convergence is therefore
of the order of the number of sections of the structured mesh along x.

3.2.2. Heat transfer

The heat transfer formulation is standard:
Z Z Z
oN k
qC½vi  ðrT Þi   N 0k dX  kðrT Þi  dX þ ðUcond þ Uf ÞN k dS ¼ 0. ð25Þ
XðtÞ XðtÞ oxi oXðtÞ

In the incremental model, a standard Galerkin formulation is used, i.e. N 0k ¼ N k . For the steady-state model, due to the
very high elemental Peclet number, a Streamline-Upwind Petrov–Galerkin (SUPG) technique is introduced, by which
N 0k 6¼ N k [36]. Based on the structured mesh described above, the biased shape functions N 0k of the advection term are com-
puted by the following formulae [37]:

3A Vk  rk
N 0k ¼ N k þ   ð1  n21 Þ  ð1  n22 Þ  ð1  n23 Þ;
4C kVk k  krk k
  ð26Þ
vi .li Pei 2 1 X3
Pei ¼ ; ci ¼ cth  ; C ¼ c1 þ c2 þ c3 ; A ¼  c  Pei
a 2 Pei 2 i¼1 i

Pei is the elemental Péclet number in direction i (element length li, central velocity vi), Vk is a nodal velocity and rk the
vector between node k and the centre of the element (reduced coordinates n1 = n2 = n3 = 0).

3.3. Examples

3.3.1. 2D comparison of FEM, SM and experiments

What are the respective capabilities of 1D models (SM) presented in Section 2.3.1 and FEM? Only experiments can tell.
Global measurements (force, torque, forward slip) are indicative, but local measurement of stresses is preferable for a full
confrontation. The latter have been obtained, starting in the 1930s, by the embedded pin technique [38] (see also [39], or
[40] for 3D measurements). Delicate as these experimental techniques may be, they have proved priceless for fine under-
standing of the mechanics of cold rolling, in 2D first then in 3D. Their quantitative results must certainly be regarded with
caution, especially for the tangential stresses, but their qualitative description of the bite has proved quite reliable, in par-
ticular through comparison with FEM (see [18] e.g. among many others).
With the above-described 3D FEM model, a 2D simulation can be done: it consists in placing two symmetry planes at
y = 0 and y = 1, with one single element along y. It can be directly compared with the SM and with experimental measure-
ments. Three experiments are modelled; they vary in their L/h ratio.
Fig. 5c, f show that SM is perfect for large L/h ratios (>3), due to the rather z-independent strain rate field – and
similarly for the stresses. On the contrary, in both cases where L/h < 3, the FEM largely outperforms the SM which
is unable to capture heterogeneities. It is also interesting to notice that the deformation occurs by localized slip bands
(reminiscent of some Slip Line Fields, or also discontinuous Upper Bound concepts): the higher L/h, the larger the num-
ber of bands, but the less sharp they are. To the emergence of each band corresponds a pressure peak. Thus, at low L/h,
a low number of strain rate concentration bands gives well-defined pressure peaks; they have nothing to do with a ‘‘fric-
tion hill’’, since the neutral point (see the tangential stress profiles) falls between the peaks! When L/h increases, the num-
ber of very faint strain rate concentration bands and the correspondingly small stress oscillations are dissolved within the
friction hill.
To conclude on these and many other numerical experiments with a practical view, the L/h > 3 criterion shows that SM
models remain useful for most cold rolling operation (with the exception of steel temper-rolling, where L is small due to
very low reduction), and even the last stands of finishing mills. They can be easily coupled with roll deformation models
[41], with advanced friction and lubrication laws (see Section 5). They have even been used with success for 3D modelling of
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6615

Fig. 5. Comparison of 2D FEM and experimental results from [38]—3 cases varying by the L/h ratio. Left: experimental contact stresses rn (‘‘sig-n’’) and
s (‘‘Tau’’) recorded by embedded transducers, compared with the Slab Method (SM) and 2D FEM (this work). Right: generalized strain-rate e_ (s1).

roll deformation, profile and flatness defects [42,43] using a series of SM’s at different y-positions. In all other cases (L/
h < 3), their neglecting shears and z-variations make them miss a part of the mechanics of the bite.

3.3.2. 3D comparison of VP, EVP and EP constitutive models

This 3D example is pictured in Fig. 6. A 1400 mm wide · 4 mm thick strip is hot rolled down to 3 mm thickness (this
corresponds to the last stand of a Hot Strip Mill). The rolls (600 mm diameter, 1600 mm barrel length) are assumed rigid
in the simulation; to mimic roll bending effects on stress patterns and profile defects, they have been given a negative (con-
cave) parabolic crown of 50 lm. Zero strip tensions are imposed. Two planes of symmetry (Oxz and Oxy) limit the sim-
ulation to 1/4th of the system. The strip mesh has 3 hexahedra in the half-thickness and 16 in the half-width with geometric
refinement near the edge. 21 elements scan the bite in RD, with refinement at entry and exit. The mesh is de-refined before
and after the bite; note that the mesh extends in both +x and x directions about twice the total strip width (2800 mm),
whereas the bite length is about 18 mm. All these conditions have been found necessary for the sake of precision (a rea-
sonable convergence in the FEM sense). Fig. 6a and b compare VP and EP behaviours (the stress-strain relations are given
in the caption). The important result pointed here is the very different out-of bite stress pattern rxx(y). In both cases, very
large stresses are induced with tensions in the centre and compression near the edges; this is due to the larger edge elon-
gation due to the negative crown. But the stress patterns are quite different:
6616 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

Fig. 6. Hot strip rolling, rigid rolls with a negative

pffiffiffi (concave) crown pffiffiffi giving a larger reduction at edges: comparison of 4 constitutive equations:
pffiffiffi (a) top view,
stress rxx, viscoplastic behaviour: r  ¼ 40  3  ðe þ 0:5Þ0:1  ð 3  e_ Þ0:2 ; (b) top view, stress rxx, elastic–plastic behaviour: r
 ¼ 120  3  ðe þ 0:5Þ0:1 ;
(c) rolling force distribution at the strip/work-roll contact. (d) friction hill p0(x, y) for the elastic–plastic (EP) case.

• In pure VP (Norton–Hoff behaviour), large stresses decay rapidly, within 1–1.5 m along +x and x; this is due to visco-
plastic relaxation by creep (inducing local thickening and/or widening), completed within Dx  w. Practically identical
results are obtained in EVP without a yield stress (K0 = 0 in Eq. (16)), which means that elasticity makes no difference
here.
• In EP, the stresses do not decay and stabilize morepquickly, they remain blocked and would lead here to flatness defects.
Choosing an EVP model with a yield stress K0 Æ 3 equal to 10% p of the total equivalent stress r
 results in ‘‘blocked’’
stresses, but an order of magnitude smaller, in proportion of K 0  3= r.

Note that the global tension force F2 = 0 is imposed here, which preserves the stress pattern rxx(y) down to the end of
the meshed domain; had the usual condition rxx = 0 been imposed at each nodal point of the section, an artificially per-
turbed zone of length w would appear at the downstream boundary of the meshed domain.
It is interesting to point out that almost identical stresses are found in the vicinity of the bite upstream and downstream;
practical observations confirm that flatness defects occurring at a given stand develop both on the entry and exit sides.
Fig. 6c gives the rolling force distribution. Again, pure VP and EVP with K0 = 0 give practically the same distribution,
with a maximum in the more deformed area near the edge. Adding a threshold (K0 5 0 corresponding to approximately
10% of the total yield stress) is intermediate, whereas a very strongly heterogeneous pattern is found in the EP case. As
shown in Fig. 6d, the 2D stress map (‘‘friction hill’’) shows a quasi-complete depletion near the centreline, due to the feed-
back of the very high out-of-bite tension stresses as boundary condition on the bite stresses. Remember this occurs with
zero global tension force.
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6617

As a conclusion on this example, the difference in the stress patterns is large; it is due not so much to the strain rate
sensitivity (the essence of VP) or to elasticity, but to the presence or not of a yield stress. EP or EVP with a dominant yield
stress is of course a necessity for cold rolling. The practice is more diversified in hot rolling modelling. EVP formulations
with a yield stress, as included in the model presented above, should be used as it opens all the possibilities for the user; the
question of the representativity of pure VP without a yield stress and its consequences on stress patterns and flatness pre-
dictions remains open.

4. Roll deformation and coupling

Roll deformation is responsible for the very cumbersome and important profile defects (along-width thickness variations
h(y)) and flatness defects; on large production mills, it is counter-acted by a number of different actuators (roll stack design,
roll bending jacks, translatable CVC rolls, . . .) [44]. Regulation loops based on continuous measurement of the strip thick-
ness profile control these actuators, which requires estimates of their mechanical effects. These are typically obtained as
local linear regressions (‘‘on line models’’) of the results of intensive campaigns of numerical simulation plans.
Roll deformation thus appears an essential step in the global modelling of the rolling process. It can be computed by a
full 3D thermo-elastic FEM [45,46]; the size of the system makes it a very costly task, but, provided fine enough roll meshes
are used, it may be considered to give reference solutions against which analytical models can be checked. Others have used
combination of semi-analytical and FEM roll modelling [47]. Here, a complete semi-analytical model is described, based on
elaborate elastic models (beam bending, Hertz contact theory), following the principles described in [48] and [49], but with
corrections based on elastic FEM of single roll deformation. The equations are discretized by influence function techniques
and framed into a single set of nonlinear equations by writing the global equilibrium (forces and moments) of the roll stack
system under relevant boundary conditions, among which the profile and flatness actuators, and contact constraints
between rolls. The roll stack (Fig. 7) is modelled as a pile of axisymmetric objects (not exactly cylinders due to the grinding
and thermal crown) in partial or total contact; a total screwdown force F tot is applied on the outermost roll chocks; inter-
roll bending forces F rb are applied to control roll deformation and therefore strip shape. These known external forces are
partitioned by the system into roll/roll and work roll/strip contact stress distributions pk(x, y) (on occasion replaced by
their x-integral, the lineic force f k(y)), to be determined (k stands for line k between rolls k  1 and k). Resulting rigid body
movements and roll deformations are to be determined as well.

4.1. Elastic roll deformation model

4.1.1. Continuum equations

Berger et al. [50] gives an expression for the 3D elastic deformation (vertical displacement Dz of surface points) of a
circular beam of length L and radius R, submitted to a net lineic load q(y) concentrated near a generator (the strip/work
roll contact or the work roll/back-up roll contact). It is the sum of a bending term and a flattening term:

• ub is the beam bending term (displacement of the centreline), including transverse shear effects. It depends on the net
force per unit length q(y) = f k(y)  f k1(y), not on the full p(x, y) distribution, due to the small ratio of the contact
arc length lc to the roll radius R (Saint-Venant’s principle). It is the solution of the usual ODE:
d4 ub q 7 þ 6mr  2m2r 1 d2 q
4
¼   . ð27Þ
dy Er  I 3ð1 þ mr Þ Er  Ar dy 2

Ftot/2 Ftot/2

Roll 3 (BUR)
f2
Contact line 2

Roll 2 (shiftable IR)

Contact line 1 f1
Roll 1 (WR)
Contact line 0 f0
Fwrb/2 Fwrb/2
y

Fig. 7. Force balance on the rolls.

6618 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

• up is Poisson’s ratio effect (Mf is the moment of the distribution q(y)):

M f ðyÞ
up ¼ 2mr . ð28Þ
E r  Ar
• Flattening uf (shortening of a radial line) can be determined on the generator loaded by p(x, y) (ufm) and on the opposite
one (ufs):
Rrzz1 ð0; y; RÞ
ufm ðx; yÞ ¼ Dz1 ðx; y; 0Þ  Dz1 ð0; y; RÞ þ K 1  ;
Er
ð29a; bÞ
Rrzz1 ð0; y; RÞ
ufs ðx; yÞ ¼ Dz1 ð0; y; DÞ  Dz1 ð0; y; RÞ þ K 2  ;
Er
in which Dz1(x, y, z) and rzz1(x, y, z) are displacements and vertical stresses derived from Boussinesq’s formula for an
arbitrary loading p(x, y) on an infinite elastic half-space (a cylinder with R = 1):
Z
3 z3
rzz1 ðx ¼ 0; y; zÞ ¼ pðx0 ; y 0 Þ 5 dx0  dy 0 ; ð30aÞ
2p R  r2 
Z
1 þ mr z 1
Dz1 ðx; y; zÞ ¼ pðx0 ; y 0 Þ 3 þ 2ð1  mr Þ  dx0  dy 0 ð30bÞ
2pEr R r r
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
with r ¼ ðx  x0 Þ2 þ ðy  y 0 Þ2 þ z2 .

7  25mr  168m2r  144m3r

K1 ¼ ;
96ð1 þ mr Þ
ð31a; bÞ
41  25mr þ 72m2r þ 48m3r
K2 ¼ .
96ð1 þ mr Þ
The total surface displacement on a given generator k 0 of roll k (k 0 = k  1 or k) is given by
0 0 00
Dzk ¼ ukb þ ukp þ U k þ ukfm þ ukfs þ U kRBM . ð32Þ
0 0
Both ufm and ufs appear since the flattening on line k 0 depends on the load f k ðyÞ (or pk ðx; yÞ) on the considered generator
0 00
via ukfm , but also on the load on the opposite generator k00 of the same roll via ukfs  U k is the bending term due to the
concentrated reaction F on the roll axes balancing the q’s. U kRBM is the Rigid Body Motion of roll k (translation due
to the screwdown/gauge control system and rotation from possibly unbalanced reaction forces between operator’s
and drive sides).
The above formulae do not account for end effects (the sudden drop in diameter at the roll barrel/axis transition). Cor-
rections have been introduced based on FEM analyses on a single roll [27,51]:
• The bending term is augmented by the following expression:
"  2 #    
F Wa Ra W =2  y W =2  y
Dub ¼ 0:608  1  exp   cos ; ð33Þ
E r  R2 R 0:418R  0:179Ra 0:260R  0:167Ra

where F is the total force on the considered roll, Wa the distance between the barrel end and the application of the force
on the axis, Ra the axis radius.
• The flattening term is shown numerically to increase due to the local drop in stiffness; on the contrary, the formulae
above, established for infinitely long cylinders, give a fall of uf at this position. This has been compensated for by a vir-
tual distribution f 0 = CFEC Æ f, symmetric of f(y) with respect to the table end. For a square-ended roll e.g.:
 
Ra
C FEC ¼ 3:12 1  . ð34Þ
R

4.1.2. Discretization by influence coefficients

These force (stress)–displacement relations are discretized by the influence function technique. This means that each con-
tact surface line Sk is meshed in y, giving a set of points y kj . Based on this grid, any contact stress distribution can be decom-
posed using linear shape functions uj(y) defined on interval [yj1, yj+1]:
X
pk ðx; yÞ ¼ fjk /j ðxÞ  uj ðyÞ ð35Þ
j

describes whenever necessary the distribution of fjk along the arc of contact of line k, i.e. for x 2 ½0; lkc . Take for
fjk /j ðxÞ
instance the bending contribution solution of Eq. (27), which depends on q(y) only. The load writes
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6619
X
qk ðyÞ ¼ qkj  uj ðyÞ ð36aÞ
j

with
y  y j1
uj ðyÞ ¼ for y 2 ½y j1 ; y j ;
y j  y j1
y jþ1  y ð36bÞ
uj ðyÞ ¼ for y 2 ½y j ; y jþ1 ;
y jþ1  y j
uj ðyÞ ¼ 0 for y 62 ½y j1 ; y jþ1 .

The solution is therefore obtained as the superposition of hand-calculated elementary solutions ui ¼ bij of Eq. (27) with the
simple load distribution q(y) = uj(y), so that the full solution is
X
ub;i ¼ ub ðy i Þ ¼ bij  qj . ð37Þ
j

bij’s are the influence coefficients in bending, which represent the effect of a unit load at yj on the displacement at yi. All
coefficients for all rolls are gathered in matrix B.
A similar treatment is performed on each displacement term. For flattening terms, the x-dependence of the load men-
tioned in Eq. (35) must be developed in some cases. For instance, in a roll/roll contact, the elliptical stress distribution and
the arc length are given by Hertz’ contact theory:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi
4 2x
/ðxÞ ¼ 1 ;
p  lc ðyÞ lc ðyÞ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  1 ð38a; bÞ
4  R  f ðyÞ 1 1 1  m2r1 1  m2r2
lc ðyÞ ¼ 2  R ¼ ðR1 1 þ R2 Þ E ¼ þ .
p  E Er1 Er2
For the strip/work-roll contact, the length and distribution are the results of the strip plastic deformation computation and
cannot be prescribed a simple, definite shape.
In the computation of influence coefficients in flattening, extensive use is made of simplifications brought by Saint-
Venant’s principle. Just as the bending terms ub(y) (and Dub, up) are calculated with the concentrated force q(y) only,
the terms rzz1(0, y, R) and Dz1(0, y, D), Dz1(0, y, R) describe the action of p(x, y) ‘‘far’’ from its application point at
z = 0 and can therefore be computed using fj only (i.e. /j(x) = d0(x)).
The term Dz1(x, y, 0) is more local and requires a full description of p(x, y). The most simple case is for the roll/roll
contact. The literal computation of integrals in Eqs. (30a) and (30b) is not possible with Eqs. (35), (38a) and (38b), but
/(x) can be decomposed into its average and an elliptical correction:
0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
 2ffi
1 4 2x 1
/ðxÞ ¼  vðxÞ þ @ 1   vðxÞA ð39Þ
lc p  lc ðyÞ lc ðyÞ lc

In the first term v(x) is the characteristic function of interval [lc/2; lc/2]. The first term, introduced in Eq. (35), leads to
easy literal integration of Eqs. (30a) and (30b). The second term has zero resultant; therefore (again according to Saint-
Venant’s principle) its contribution needs be computed only in the close vicinity of yj, i.e. on [yj1, yj+1]. Using Eqs.
(35), (38a) and (38b):
1  m2r
Dz1 ð0; y j ; 0Þ ¼ ð2  Ln2  1Þ  fj ð40Þ
pEr
For the work roll/strip contact, the contact stress map is more complex and cannot be described by a formula. The decom-
position of p(x, y) then writes:
X
p0 ðx; yÞ ¼ p0ij /i ðxÞ.uj ðyÞ ð41Þ
j

where /i(x) is a linear shape function on [xi1, xi+1] similar to uj(y). Furthermore, the required displacement field is now a
function of x and y, Dz1(xk, yl, 0). The analytical integration of Eq. (30b) with Eq. (41) is now possible, and allows to write
down the 4-subscript influence coefficients sought. Again, a decomposition of p0(x, y) into its x-average p0 ðyÞ and a com-
plement p0 ðx; yÞ  
p0 ðyÞ with zero resultant allows to replace it by the computation of less numerous 3-subscript coeffi-
cients, except in a vicinity of yj to be chosen (a few times the contact arc length).
6620 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

4.1.3. Global matrix solution

The equilibrium equation of each roll k (comprising generator k  1 and k) writes (Fig. 7)
X
ðfjk1  fjk Þ ¼ Ek . ð42Þ
j

where Ek is the (known) balance force applied on a roll (screwdown force for the outermost rolls of the stack). Writing
fjk1  fjk ¼ qkj , the net bending force distribution on roll k, and gathering all the f ’s, q’s and E’s of all rolls in supervectors
F, Q and E, Eq. (42) can be written [27,51]:
T  Q ¼ T  S  F ¼ N  E; ð43Þ
where T, S and N are assembly matrices (made out of 0 and ±1).
The other equations to be written are the compatibility of displacements at each roll/roll contact line k between roll k
and k + 1 (including line k = 0 for the lower/upper work roll contact in case of roll kiss):
0 0 00
Dzk þ bk þ bkþ1 ¼ ukb þ ukp þ U k þ ukfm þ ukfs þ U kRBM þ bk þ bkþ1 ¼ 0 ð44Þ

where b’s are the crowns (grinding + thermal) of rolls k and k + 1.

As above, the bending displacements at all nodes i of all lines k 0 , due to qj (i.e. ub + up) and to concentrated forces Ek are
written together with the Rigid Body Motions U kRBM as a global vector U, the displacements due to flattening (ufm, ufs) give
vector G, the rolls crowns form vector B. The influence coefficient formulation described above gives:

U ¼ B  Q þ J  E þ O  W;
ð45Þ
G ¼ KðFÞ  F;

where B, J, K are matrices of influence coefficients, respectively for bending forces q, overall forces Ek and flattening terms
(f, p). O is an assembly matrix for Rigid Body Motions collected in vector W. Eq. (44) writes
L  U þ M  B þ G ¼ 0; ð46Þ
where L and M are again assembly matrices.
All this finally results in a matrix equation summarizing Eqs. (43) and (45):
     
LBSþK LO F L  J  E  M  B
 ¼ ; ð47Þ
T S 0 W N E

where the unknowns are the force distribution on each contact line, gathered in F, and the Rigid Body Movements of all
rolls W.
Eq. (47) is indeed nonlinear, first because flattening influence coefficients in K depend on the load solution F via the arc
lengths lc, and also because the contact areas between crowned rolls are not known at first—they depend on the loads as
well. A Newton–Raphson procedure is therefore used.
In case of roll kiss (see Fig. 9), the global force F tot is NOT known because the ‘‘kissing force’’ between the work rolls
must be added to E. This kissing force is an unknown of the problem expressed as P Æ F where the assembly matrix P col-
lects only those fj0 corresponding to work roll/work roll contact areas defined by dj = 0 (d is the distance between the hor-
izontal symmetry axis and the work roll profile). Finally, the system becomes
     
L  B  S þ LJP þ K LO F L  J  E  M  B
 ¼ ð48Þ
T S 0 W N E

and the condition "j, dj P 0 is added to the convergence criteria. The problem is fully determined by assuming that a per-
fect AGC (Automatic Gauge Control) ensures exactly and instantaneously the target strip thickness on the centreline.
Finally, from the (F, W) solution of Eq. (47) or (48), all the unknown force distributions, contact pressures, displace-
ments can be derived by the equations above.

4.1.4. Conclusion
Complex as they may seem, these developments are worthwhile, since they replace a super-intensive computation of an
enormous elastic system by a module that computes its deformation with a high precision within a fraction of a second, a
time negligible compared with the strip thermo-mechanical model [27]. Good agreement has been found with other models
[27] and with experiments [27,51,52].
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6621

4.2. 2D/3D steady-state roll thermal model

Thermal crown is an integral part of Eq. (44); moreover the heat transfer between the strip and the roll is an essential
part of the Hot Rolling Process, which depends strongly on the thermal field within the rolls. It is considered here that only
the temperature of the work roll is of importance, the back up rolls are included in the analysis only as flux extractors, i.e.
part of the boundary conditions.
Causes of departure from steady state from a mechanical point of view have already been mentioned in Section 2.2. Roll
thermal evolution and roll wear along one mounting of the WR are other ones. In hot strip rolling in particular, the dis-
continuous, coil-to-coil character of the process, with changing formats, results in time-variations of the thermal field in the
WR. This can be included in the modelling by considering the heat flux as a variable determined by the rolling schedule
[27]. Here, for the sake of conciseness, we discuss only the time-independent model. Based on the work of [53] e.g., it is
considered that a skin layer exists in the roll where noticeable high-frequency perturbations are induced by boundary con-
ditions (heat transfer from the hot strip, cooling by water jets, . . .) so that the temperature is T(r, h, y); the thickness dT of
the skin is taken as
pffiffiffiffiffiffiffiffiffiffiffiffi
dT  4  ar =xr ð49Þ
where ar is the thermal diffusivity of the roll material and xr the angular velocity.
Farther inside the roll, the perturbations are smoothed by the time necessary for the thermal information to reach this
zone. Therefore, T does not depend on h anymore: T(r, y).
This is the basis for a considerable simplification of the model (Fig. 8). The roll is divided into a single core model with
Tcore(r, y), and a skin model where furthermore heat fluxes along y are supposed small, so that it can be represented as a
series of 2D FEM at ordinates yj, T(r, h, yj) = Tskin,j(r, h). The necessary coupling between the skin and the core is provided
by the flux exchanged between them at the interface:
 core   skin;j 
int oT ðr; yÞ oT ðr; #Þ
Uj ¼ k r j ¼ kr
or y or
(the skin is taken thick enough that at the interface, T and oT/or do not depend on h). From linearity, it is inferred that,
using the skin models:
Uint ¼ C T þ DT  T int . ð50Þ
Similarly, in the core model:
T int ¼ AT þ B T  Uint ; ð51Þ
where (T int, Uint) are vectors gathering the interfacial temperatures and fluxes in the N sections j. Influence coefficients
matrices BT and DT are determined at the beginning by N + 1 FEM computations. Then from Eqs. (50) and (51):
1
T int ¼ ðI  B T  DT Þ  ðAT þ B T  C T Þ ð52Þ
which can then be introduced as the boundary condition for the determination of the temperature fields in both skin and
core models, which are treated by the SUPG technique described in Section 3.2.2. Finally, the thermal crown to be intro-
duced into terms bk in Eq. (44) is obtained by a FEM solution of the standard linear thermo-elastic constitutive equations:
r ¼ 2lelr  e þ kelr  TrðeÞ  I  aT ;r  ð2lelr þ 3kelr Þ  ðT  T 0 Þ  I. ð53Þ

Fig. 8. Principle of the thermal model: coupling of skin/core 2D FEMs.

6622 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

This series of 2D FEM remains much less costly than a full 3D simulation of the roll thermal regime. It has been verified
[27] that a very good agreement with closed-form solutions of the literature is obtained.

4.3. Example: Thin strip rolling and roll kiss

In Fig. 9, some results are given for thin strip rolling on a 4-high mill. The dimensions are described in the figure. A pure
VP behaviour has been assumed to improve convergence, with:
pffiffiffi  0:21 pffiffiffi 0:13
 ¼ 520  3  104 þ e
r  3  e_ . ð54Þ

Friction is modelled by a ‘‘Norton’’-type friction:

pffiffiffi 0:21 0:1
s ¼ 156  3  ð104 þ eÞ  ðv2  v1 Þ . ð55Þ
The strip–roll coupling computes a severe roll deformation, including bending and indentation, as shown by the ‘‘active
generator’’ in Fig. 9d; due to the small strip thickness, roll-kiss occurs (contact of the upper (pictured) and lower (not
shown) work-roll) outside the bite. Note the refined meshes of roll (influence functions) and strip (FEM) near the edge.
Due to the large L/h ratio, the z-gradients are negligible so that only one element has been used in the thickness. The peak
of the load distribution is found near the edge (see the 2D friction hill in Fig. 9b). Fig. 9c displays the deformed roll shape
in a longitudinal cross-section (centreline).

Fig. 9. A strip–roll coupled 3D computation of thin strip cold rolling with roll kiss: (a) rolling mill description. The strip is 0.4 mm thick · 400 mm wide,
reduction is 40%. (b) the 2D-friction hill (strip/work-roll normal contact stress); (c) longitudinal cross-section showing the roll deformed profile;
(d) transverse cross-section in plane x = 0.
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6623

5. Conclusion and complements

The developments described above point to the peculiarities of rolling processes and their consequences on modelling
strategies. First, even in shape and profile rolling, the shapes are never as intricate as some complex forgings or castings;
the strain heterogeneities and distortions are moderate; this is why remeshing in FEM is practically never required. On the
other hand, the coupling of the roll and rolled stock deformation is an essential feature of the process, which has been
addressed ever since the 1930s in 2D and the 1960s in 3D. Several techniques have proved valuable, the most efficient
in terms of cost/quality ratio remains the influence function method. The important point here is that the essential feature
of a model is the intensity of the coupling between the two models [54,55,27], to be tuned to the intensity of the physical
coupling between both deformations.
Another important feature is the overwhelming importance of the steady state stages, although transients may also have
their own importance. Thus, rolling modelling has been the field of exercise of Eulerian, quasi-Eulerian, ALE techniques of
various kinds in solid mechanics. One thing to remember is that in such (quasi)-Eulerian techniques as described in this
paper, the extent of the domain considered is crucial: a kind of experimental Saint-Venant’s principle states that the extent
of the mesh on either side of the roll bite must be larger than, say, twice the greatest dimension of the cross-section to avoid
any spurious impact of the stress condition at extremities (homogeneous tension stress). The demand may however become
less stringent if alternative boundary conditions are chosen, such as an imposed integral of the stress—the tension force—
letting the stress distribution establish as it should, which is more realistic but leads to more algorithmic complexity (e.g. one
more level of iterations).
With these peculiarities, mass production steel and aluminium rolling processes were certainly the first example where
modelling (3D FEM in the forefront) went beyond the simple understanding of the thermo-mechanics of the processes, and
participated early in the design, set up and automatic control of the production lines. This is however not true for other
metals and alloys, the production of which remains much closer to the craft it used to be and owes little to modelling even
today.
Beyond the thermal/mechanical coupling, insertion of some physics in the rolling models has been very active for a long
time. This is not the place for detailing the predictive microstructural models which are by no means specific of rolling (see
the paper by Logé and Chastel in the present issue [56], and references therein). It is worth however mentioning tribological
models, since friction is the driving force of rolling processes (rolling with too low friction leads to skidding and instability).
2D modelling of rolling with coupled HydroDynamic (HD) models began in the 1960s [57]. Shortly after, HD was replaced
by more realistic mixed lubrication models [58–60] which themselves coupled a thin-film fluid mechanics approach (Rey-
nolds’s equation) with a microplastic analysis of asperity interaction and real area of contact [61,62]. Coupling has been
performed in 2D, slab models [57–65] and predicts such variables as lubricant film thickness, real area of contact and final
strip roughness, local friction coefficient, on top of the usual stress, strain and temperature fields, force and torque. These
however rely on the (still arbitrary) choice of a dry or boundary microscopic friction coefficient, which limits the predic-
tivity of such models. More understanding of the mechanical and tribological properties of the 2D and 3D reactive bound-
ary films formed during friction at the microscale will be necessary before this state of affairs may change. Coupling such
enhanced models with 3D rolling modelling, providing a better description of local friction, might significantly improve all
the force distribution and roll deformation prediction, and help using such profile and flatness actuators as differential
lubrication or differential roll cooling.
Among the open problems is flatness prediction. 3D models easily predict the relative differential elongation of the
strip along DT, expressed in U.I. = 105, from the incoming profile and the exit profile (i.e. the roll profile under load).
The relation between the U.I. profile and the strip flatness is a matter of experience. It was attempted to predict explic-
itly the flatness (i.e. out-of-plane movement due to buckling of highly compressed lines) with the coupled rolling model.
It must first be realised that flatness defects are convective (buckling) instabilities which are created by high compressive
stress at both bite ends and are afterwards transported by the material velocity. Therefore, they cannot be captured by
steady-state models. Attempts with incremental computations failed also [66]: the large and flat hexahedra used to
cover the very large area outside the roll bite proved too stiff and unable to buckle. The remedy is to proceed in
two steps: first, a strip/roll coupled computation is performed under steady state hypothesis, with a horizontal as well
as a vertical symmetry plane to avoid geometrical instabilities. The out-of-bite state of stress is then extracted and
either transferred to a structural mechanics FEM, using shell elements; or used as an entry of a software which looks
for the optimum parameterized shape which minimizes the elastic energy of the free strip [66,67]. These techniques pro-
duce quite realistic results, but they fail to capture the feedback of the buckling phenomenon on the out-of-bite state of
stress, i.e. on the rxx boundary conditions in the near-vicinity of the bite which, in severe cases, may impact on the
vertical contact force distribution f(y) (similar to Fig. 6d) and therefore on the roll profile under load, i.e. ultimately
on flatness itself. The one-step, fully predictive coupled modelling of flatness therefore remains desirable, albeit a for-
midable challenge which might be won if 3D elements are used in the bite vicinity and coupled with shell elements
farther off.
6624 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625

The shape of the strip may also be severely impaired by mill vibrations known as chatter, another challenge for future
work. In particular, vibrations in the 100–200 Hz range, known as ‘‘third-octave chatter’’ are particularly detrimental, as
they may lead to large amplitude fluctuations of strip thickness and to strip breaks; ‘‘fifth-octave chatter’’, in the 500–
600 Hz range, leads to periodic marks on the surface and thickness variations, although of a smaller amplitude. The only
known efficient remedy is to decrease the rolling speed when chatter sets in. Although mill vibrations have been studied
experimentally in 3D, their coupling with strip deformation can be imagined in 2D only, using non-steady-state slab models
accounting for the temporal gap thickness variations imposed by the vertical vibration of the rolls (e.g. [68,69]). It has
indeed brought rich information on the origins and mechanisms of chatter. In particular, the self-excitation of the mill
due to the phase quadrature between the roll movement and the induced front and back tension oscillations is well estab-
lished. Such models are able to describe how an incipient small perturbation (strip thickness, . . .) may be amplified and
which is the dominant resulting frequency. The relationship with friction is also explored; the interesting thing is that both
excessive and insufficient friction seem to provide favourable conditions for chatter to occur.

References

[1] T. von Karman, Beitrag zur Theorie des Walzvorganges, ZAMM 5 (1925) 139–141.
[2] E. Orowan, The calculation of roll pressure in hot and cold flat rolling, Proc. Instn. Mech. Engrg. 150 (1943) 140–167.
[3] K. Mori, K. Osakada, Simulation of 3D deformation in rolling by the FEM, Int. J. Mech. Sci. 26 (9–10) (1984) 515–525.
[4] K. Yamada, S. Ogawa, S. Hamauzu, T. Kikuma, 3D analysis of mandrel rolling using RP-FEM, in: E.G. Thompson et al. (Eds.), Proc.
NUMIFORM 89, Fort Collins, USA, A.A. Balkema, Rotterdam, 1989.
[5] J.-L. Chenot, P. Montmitonnet, A. Bern, C. Bertrand-Corsini, A method for determining free surfaces in steady state finite element computations,
Comput. Methods Appl. Mech. Engrg. 92 (2) (1991) 245–260.
[6] Y.S Lee, P.R. Dawson, T.B. Dewhurst, Bulge prediction in steady state bar rolling process, in: E.G. Thompson et al. (Eds.) Proc. NUMIFORM 89,
Fort Collins, USA, A.A. Balkema, Rotterdam, 1989.
[7] M. Vacance, Modélisation tridimensionnelle par éléments finis du laminage à chaud des tubes sans soudure (3D FEM modelling of seamless tubes hot
rolling), PhD Dissertation, Ecole des Mines de Paris, France, 1993 (in French).
[8] E.N. Dvorkin, M.B. Goldschmit, M.A. Cavaliere, P.M. Amenta, O. Marini, W. Stroppiana, 2D finite element parametric study of the flat rolling
process, J. Mat. Process. Technol. 68 (1997) 99–107.
[9] H.J. Huisman, J. Huétink, A combined Eulerian–Lagrangian 3D FEM analysis of edge-rolling, J. Mech. Working Technol. 11 (1985) 333–
353.
[10] D.R. Bland, H. Ford, An approximate treatment of the elastic compression of the strip in cold rolling, J. Iron Steel Inst. 171 (1952) 245–249.
[11] J.M. Alexander, On the theory of rolling, Proc. R. Soc. London A 326 (1972) 535–563.
[12] P. Cosse, M. Economopoulos, Mathematical study of cold rolling, C.N.R.M. 17 (1968) 15–32.
[13] P.A.F. Martins, M.J.M. Barata Marques, Upper bound analysis of plane strain rolling using a flow function and the weighted residual method,
Int. J. Numer. Methods Engrg. 44 (1999) 1671–1683.
[14] C. Counhaye, Modélisation et contrôle industriel de la géométrie des aciers laminés à froid (modelling and industrial control of the geometry of cold
rolled steel), PhD Dissertation, University of Liège, Belgium, 2000 (in French).
[15] A. Orban, D. Steinier, C. Counhaye, Shape control in cold rolling, in: Proc. 6th Int. Rolling Conf. (METEC), 1994, pp. 257–265.
[16] H.C. Lee, S. Kobayashi, New solution to rigid-plastic deformation problems using a matrix method, ASME J. Engrg. Ind. 95 (1973) 865–873.
[17] G.D. Lahoti, S.N. Shah, T. Altan, Computer-aided analysis of deformations and temperatures in strip rolling, ASME J. Engrg. Ind. 100 (1978) 159–
166.
[18] G.J. Li, S. Kobayashi, Rigid-plastic finite element analysis of plane strain rolling, ASME J. Engrg. Ind. 104 (1982) 55–64.
[19] A.C.W. Lau, R. Shivpuri, P.C. Chou, An explicit time integration elastic–plastic finite element algorithm for analysis of high speed rolling,
Int. J. Mech. Sci. 31 (7) (1989) 483–497.
[20] S.W. Xiong, J.M.C. Rodrigues, P.A.F. Martins, Application of the element-free Galerkin method to the simulation of plane strain rolling,
Eur. J. Mech. A/Solids 23 (2004) 77–93.
[21] R. Deltombe, S. Huart, M. Dubar, L. Dubar, From the iron fine production to wrenching phenomena: local cold rolling mill enhancement,
in: S. Støren (Ed.), Proc. ESAFORM 7, Tronheim, Norway, April 2004, Norwegian University of Science and Technology, 2004.
[22] J.H. Beynon, Modelling the boundary conditions for thermo-mechanical processing—oxide scale behaviour and composition effects, Model. Simul.
Mater. Sci. Engrg. 8 (2000) 927–945.
[23] B. Picqué, S. Mouret, P.-O. Bouchard, P. Montmitonnet, M. Picard, Mechanical behaviour of iron oxide scale: experimental and numerical study,
in: Proc. 3rd Int. Conf. Tribology in Manufacturing Processes, ICTMP, Nyborg, Denmark, 16–18 June 2004.
[24] H.M. Liu, J. Dan, Y.R. Wang, Strip layer method for simulation of the 3D deformations of plate and strip rolling, Comm. Numer. Methods Engrg.
20 (2004) 183–191.
[25] C. Liu, P. Hartley, C.E.N. Sturgess, G.W. Rowe, Analysis of stress and strain distributions in slab rolling using the elastic–plastic finite element
method, in: K. Mattiasson, A. Samuelsson, R.D. Wood, O.C. Zienkiewicz (Eds.), Proc. NUMIFORM 86, Gothenburg, Sweden, 1986, A.A. Balkema,
Rotterdam, 1986.
[26] G. Sola, Contribution à la modélisation thermomécanique 3D par éléments finis du laminage à chaud des tubes: calcul multicage (3D FEM modelling
of hot rolling of tubes: multistand computation), PhD Dissertation, Ecole des mines de Paris, 1994 (in French).
[27] A. Hacquin, Modélisation thermomécanique 3D du laminage: couplage bande—cylindres (3D thermomechanical modelling of rolling processes:
strip/roll coupling), PhD Dissertation, Ecole des Mines de Paris, 1996 (in French).
[28] A. Hacquin, P. Montmitonnet, J.-Ph. Guillerault, A steady state thermo-elastoviscoplastic finite element model of rolling with coupled thermo-elastic
roll deformation, J. Mater. Process. Technol. 60 (1–4) (1996) 109–116.
[29] P. Montmitonnet, P. Gratacos, R. Ducloux, Application of an anisotropic viscoplastic behaviour in 3D finite element simulations of hot rolling,
J. Mater. Process. Technol. 58 (2–3) (1996) 201–211.
 P. Montmitonnet / Comput. Methods Appl. Mech. Engrg. 195 (2006) 6604–6625 6625

[30] P. Noat, P. Montmitonnet, Y. Chastel, R. Shahani, Anisotropic 3D modelling of hot rolling and plane strain compression of Al alloys, in: S.F. Shen,
P.R. Dawson (Eds.), Proc. NUMIFORM 95, Ithaca, NY, USA, June 1995, A.A. Balkema, Rotterdam (1995).
[31] P. Montmitonnet, A. Hacquin, Implementation of an anisotropic friction law in a 3D finite element model of hot rolling, in: S.F. Shen, P.R. Dawson
(Eds.), Proc. NUMIFORM 95, Ithaca, NY, USA, June 1995, A.A. Balkema, Rotterdam, 1995.
[32] M. Pietrzyk, J.G. Lenard, Thermal–Mechanical Modelling of the Flat Rolling Processes, Springer, Berlin, 1991.
[33] A.A. Tseng, Thermal modelling of roll and strip interface in rolling processes. Part 1: Review, Numer. Heat. Transfer A 35 (1999) 115–133.
[34] A.A. Tseng, Thermal modelling of roll and strip interface in rolling processes. Part 2: Simulation, Numer. Heat. Transfer A 35 (1999) 135–154.
[35] A.A. Tseng, F.H. Lin, A.S. Gunderia, D.S. Ni, Roll cooling and its relationship to roll life, Met. Trans. 20A (1989) 2305–2320.
[36] A.N. Brooks, T.J.R. Hughes, SUPG formulations for convection-dominated flows with particular emphasis on the incompressible Navier–Stokes
equations, Comput. Methods. Appl. Mech. Engrg. 32 (1982) 199–259.
[37] E. Pardo, D.C. Weckman, A fixed-grid finite element technique for modelling phase change in steady-state conduction–advection problems,
Int. J. Numer. Methods Engrg. 29 (1990) 969–984.
[38] F.A.R. Al-Salehi, T.C. Firbank, P.R. Lancaster, An experimental determination of the roll pressure distributions in cold rolling, Int. J. Mech. Sci.
15 (1973) 693–710.
[39] J.G. Lenard, Friction and forward slip in cold strip rolling, STLE Tribol. Trans. 35 (3) (1992) 423–428.
[40] J. Jeswiet, Measuring metal deformation interface forces and temperatures, Trans. CSME 17 (4A) (1993) 633–645.
[41] A. Atreya, J.G. Lenard, A study of cold strip rolling, J. Engrg. Mater. Techn. (Trans. ASME) 101 (1979) 129–134.
[42] O. Pawelski, W. Rasp, J. Rieckman, Application of a mathematical model for the cold rolling process on six-high mills, in: J.L. Chenot, E. Oñate
(Eds.), Proc. Modelling of Metal Forming Processes, Sophia-Antipolis, France, Kluwer Academic Publishers, Dordrecht, 1988.
[43] J. Leven, N.G. Jonsson, O. Wiklund, J. Edberg, Strip and plate profile and edge drop in flat rolling, in: Proc. Advanced Technology of Plasticity,
Kyoto, Japan, Japan Society for Technology of Plasticity, 1990.
[44] B. Berger, E. Neushütz, G. Mücke, Experiences with a flatness control system including axial roll shifting, in: Proc. 4th Int. Steel Rolling Conf.,
Deauville, France, E13.1–E13.6, Association Technique de la Sidérurgie (ATS), Paris, 1987.
[45] S.X. Zhou, P. Funke, J. Zhong, C. Plociennik, Modification of a classical formula for determination of roll flattening in flat rolling, Steel Res.
67 (11) (1996) 491–494.
[46] T.H. Kim, W.H. Lee, S.M. Hwang, An integrated FE process model for the prediction of strip profile in flat rolling, ISIJ Int. 43 (12) (2003) 1947–
1956.
[47] J. Yanagimoto, M. Kiuchi, 3D simulation system for coupled elastic/rigid plastic deformation of rolls and workpieces in strip rolling processes,
in: J.L. Chenot et al. (Eds.), Proc. NUMIFORM 92, Sophia-Antipolis, France, A.A. Balkema, Rotterdam, 1992.
[48] K.N. Shohet, N.A. Townsend, Roll bending methods of crown control in four-high plate mills, J. Iron Steel Inst. 206 (11) (1968) 1088–1098.
[49] K. Yamada, S. Ogawa, M. Ataka, 3D analysis of flat rolling using rigid-plastic finite element method coupled with roll deformation analysis,
in: J.L. Chenot et al. (Eds.), Proc. NUMIFORM 92, Sophia-Antipolis, France, A.A. Balkema, Rotterdam, 1992.
[50] B. Berger, O. Pawelski, P. Funke, Die elastische verformung der Walzen von Vierwalzengerüste, Arch. Eisenhüttenwes. 47 (1976) 351–356.
[51] A. Hacquin, P. Montmitonnet, J.-P. Guillerault, A 3D semi-analytical model of rolling stand deformation with finite element validation, Eur.
J. Mech. A/Solids 17 (1) (1998) 79–106.
[52] A. Hacquin, P. Montmitonnet, J.Ph. Guillerault, Experimental validation of a rolling stand elastic deformation model, J. Mater. Process. Technol.
45 (1994) 199–206.
[53] J.V. Poplawski, D.A. Seccombe, Bethlehem’s contribution to the mathematical modelling of cold rolling in tandem mills, Iron Steel Eng. 57 (1980)
47–58.
[54] M.J. Grimble, M.A. Fuller, G.F. Bryant, A non-circular arc force model for cold rolling, Int. J. Numer. Methods Engrg. 12 (1978) 643–663.
[55] P. Gratacos, P. Montmitonnet, C. Fromholz, J.L. Chenot, A plane strain elastoplastic finite element model for cold rolling of thin strip, Int. J. Mech.
Sci. 34 (3) (1992) 195–210.
[56] R.E. Logé, Y.B. Chastel, Coupling the thermal and mechanical fields to metallurgical evolutions within a finite element description of a forming
process, Comput. Methods Appl. Mech. Engrg., this issue, doi:10.1016/j.cma.2004.11.034.
[57] W.R.D. Wilson, J.A. Walowit, An isothermal hydrodynamic lubrication theory for strip rolling with front and back tension, in: Proc. Tribology
Convention, Douglas, Isle of Man, UK, 12–15 May 1971, I. Mech. E., London, 1972, pp. 164–172.
[58] Y.H. Tsao, L.B. Sargent, A mixed lubrication model for the cold rolling of metals, ASLE Trans. 20 (1977) 55–63.
[59] S. Sheu, W.R.D. Wilson, Mixed lubrication of strip rolling, STLE Tribol. Trans. 37 (1994) 483–493.
[60] M.P.F. Sutcliffe, K.L. Johnson, Lubrication in cold strip rolling in the mixed regime, Proc. Instn Mech. Engrg. 204 (1990) 249–261.
[61] W.R.D. Wilson, S. Sheu, Real area of contact and boundary friction in metal forming, Int. J. Mech. Sci. 30 (7) (1988) 475–489.
[62] M.P.F. Sutcliffe, Flattening of random rough surfaces in metal forming processes, ASME. J. Tribol. 121 (3) (1999) 433–440.
[63] H.S. Lin, N. Marsault, W.R.D. Wilson, A mixed lubrication model for cold strip rolling—Part I: theoretical, STLE Tribol. Trans. 41 (3) (1998) 317–
326.
[64] N. Marsault, Modélisation du régime de lubrification mixte en laminage à froid (modelling of mixed lubrication regime in cold rolling). Ph.D.
Dissertation, Ecole des Mines de Paris, 1998 (in French).
[65] M.P.F. Sutcliffe, P. Montmitonnet, Numerical modelling of lubricated foil rolling, Rev. Mét.- SGM 98 (2001) 435–442.
[66] H. Marchand, Modélisation de la planéité en sortie de laminage des produits plats (modelling flatness in flat rolling). PhD Dissertation, Ecole des
Mines de Paris, 2000 (in French).
[67] F.D. Fischer, F.G. Rammerstorfer, N. Friedl, Residual stress-induced center wave buckling of rolled strip metal, J. Appl. Mech. (Trans. ASME)
70 (2003) 84–90.
[68] P.H. Hu, K.F. Ehmann, A dynamic model of the rolling process. Part I: Homogeneous model, Int. J. Mach. Tool Manuf. 40 (2000) 1–19.
[69] P.H. Hu, K.F. Ehmann, A dynamic model of the rolling process. Part II: Inhomogeneous model, Int. J. Mach. Tool Manuf. 40 (2000) 21–31.